We present a comprehensive analysis of the recent inter-annual variation of the atmospheric CO₂ growth rate, with a special focus on the 2002-2003 period, using a state of the art atmospheric inversion, process driven model simulations (land and ocean), and recent biomass burning estimates. The inverse estimates compare favourably well with the model simulations over North Asia and indicate a large contribution of the fire anomaly to the total anomaly, for that region in 2003. Over Europe, the spatial distribution of the inverse and bottom-up flux anomalies for 2003 have similarities but the time evolution of the total fluxes still need to be reconciled.

Results from recent models in the coupled carbon cycle climate model intercomparison project (C4MIP) indicate a positive feedback to global warming from the interactive carbon cycle, but the magnitude varies widely. A typical model simulates an additional increase of 90 ppmv in the atmospheric CO₂, and 0.6 degree additional warming due to this feedback, but some model can be as large as 250ppm. Using a liner perturbation framework, we analyze what might have caused such large discrepancy in the models, with a focus on land where the largest uncertainties lie. Change in NPP such as different sensitivity to the CO₂ fertilization effect is one where in some models it is modest largely due to the multiple limiting factors constraining terrestrial productivity and carbon loss. The large differences among the models are also manifestations of other poorly constrained processes such as the turnover time and rates of soil decomposition.

Data from long-term measurements of carbon balance in boreal, mid-latitude and tropical ecosystems are used to assess the mechanisms that drive changes in ecosystem carbon balance in response to a changing climate. We find that most model parameterizations overestimate the temperature sensitivity of ecosystem respiration and underestimate the role of soil water balance in controlling respiration and flammability. We conclude that model assessments of climate—carbon feedbacks must carefully simulate regional precipitation, evaporation, evapotranspiration, and water balance, including factors leading to fires (e.g. sources of ignition), in addition to assessing changes in temperature. Covariances among these drivers of ecosystem respiration and vegetation change may be critically important for these simulations.